Spectral Characteristics of Digit- Simulating Speech Sounds By D. P. BORENSTEIN (Manuscript received July 11, 1963) A spectral analysis has been performed on a number of spoken vowel sounds, in particular those sounds causing digit registration in a TOUCH- TONE receiver. The analysis, implemented by computer methods, provides a definitive picture of the nature of digit simulation in TOUCH-TONE calling. I. INTRODUCTION A digit simulation in TOUCH-TONE calling (Ref. 1, pp. 9-12, 15-16) is, by practical definition, a speech segment capable of causing digit registration in a TOUCH-TONE signaling system. Spectral analyses have been performed on a number of speech segments, each of which was selected solely on the basis of having the above property. Briefly, a valid TOUCH-TONE signal requires the simultaneous presence of two code frequencies for a certain minimum length of time, and with some minimum signal-to-noise ratio. It was therefore theoretically an- ticipated (Ref. 1, pp. 10-12) that each of these speech segments would be linked by two other common characteristics: (1) a frequency spectrum having two sharply dominant peaks, and (2) a high degree of periodicity for some minimal length of time. There is good reason to believe that speech segments of this general nature are likely to be troublesome in any signaling system based on the transmission of voiceband tones over speech channels. Due to the inherent rarity and relatively brief duration of the voice- produced digit simulation, some special procedures were required both in obtaining and analyzing these speech segments. The remainder of this article comprises a description of these procedures, followed by a presentation and discussion of the resulting spectral analyses. 2839 2840 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1963 10-KC FREQUENCY STANDARD Fig. 1 — Apparatus for recording digit-simulating speech segments. II. COLLECTING THE SPEECH SAMPLES The digit-simulating speech segments were obtained by recording raw speech onto magnetic tape loops with the two-track recording arrangement shown in Fig. 1. Using a GO-inch loop of tape at a speed of 15 in/sec, speech is continu- ously recorded at point A, played into a standard TOUCH-TONE re- ceiver at point B, and, if there is no receiver output, erased at point E after traversal of the loop. Simultaneously, on a second track, a 10-kc pilot frequency is continuously recorded and erased at points B and C, respectively. If at any time there is a TOUCH-TONE receiver output, indicating the presence of a digit-simulating speech segment just past point B, the timing network is triggered. The timing network then performs two operations: (1) it disables the 10-kc record and erase after a delay of 35 ms, and (2) it stops the tape transport after a delay of 2 seconds (half the loop traversal time) . This process yields a 60-inch length of tape consisting of about 29 inches each of pre- and post-simulation speech plus a 1.5-inch (110 ms at 15 in/sec) segment which contains both the actual simulating speech sample and the 10-kc pilot frequency. In this manner, fourteen such samples were obtained, at the average rate of about one per ten hours of raw speech — an indication of the extreme rarity of simulation with the present TOUCH-TONE receiver. III. ANALOG-TO-DIGITAL CONVERSION AND PRINTOUT By means of encoding equipment developed by the Acoustics Research Department, the fourteen digit-simulating speech segments were con- DIGIT-SIMILATIXG SPEECH SOUNDS 2841 verted from analog form to an eleven-bit digital signal. The sampling rate of 10 kc was gated directly from the pilot track of the original analog tape, thus eliminating sources of error due to tape flutter during the original recording process. Once the digital tape was obtained, the conversion process was reversed to obtain an accurate X-Y recording of each of the fourteen speech waveforms. Visual inspection of these waveforms, two of which are shown in Fig. 2, confirms their periodic nature (the periodicity of the samples shown in Fig. 2 would be still more evident were it not for the fact that most speech fundamentals are considerably attenuated by telephone apparatus). IV. SPECTRAL ANALYSIS The fourteen speech samples, in eleven-bit digital format, were then subjected to a "pitch synchronous" 2 Fourier analysis on the IBM-7090 computer. The pitch synchronous analysis consisted essentially of a conventional Fourier analysis performed on each successive funda- mental pitch period in the speech sample. These pitch periods, in turn, were determined on the computer by counting the number of sampling intervals (each being 100 /usee) between successive maxima in the wave- form and then interpolating between samples for greater accuracy. This method of Fourier analysis is ideally suited to waveforms that maintain an almost-periodic structure over an appreciable length of time. For each speech segment analyzed, the computer output consisted of a sequential set of bar graphs, one for each fundamental pitch period of the speech waveform. Each graph, in turn, is a plot of harmonic ampli- tude (the Euler coefficient) in db versus harmonic number. In addition, each graph gives the "instantaneous pitch" (i.e. the reciprocal of the period) of each fundamental period analyzed. Figs. 3 and 4 show the TIME IMS Fig. 2 — Analog waveforms of two digit-simulating speech segments. Also shown are the sex of each speaker and Hie particular phoneme causing the simulation. Fourier spectra of digit simulations 1 and 2 are shown in Figs. 3 and 4, respectively. Arrows indicate periodicity, with the large arrow showing the approximate start of the digital simulation. 1 (a) 170.30/ 1 1 I X X X X X X X X XX X X X X XX X XX XXI XXX X XXX X XXX xxxx X XXX xxxx xxxxx xxxx XXXXX X ■xxxx xxxxx XXXXX X XXXXX X xxxxx XXXXX X IXXXX XXXXX XX xxxxx XXXXX XX XXXXXXXI >x«xx IXXXXXXX XXXXX XXXXXXXI xxxxxxxxxxxxxx imiimixxixx xxxxxxxxxxxxxxxx IXXXXXXXXXXXXKXXX xxxxxxxxxxxxxxxxx xxxxx xxxxx xxxxx xxxxx xxxxx xixxxxxxxxx milium xxxxxxxxxxx X xxxx xxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxx IXXXXXXXXIXXXIXXXXXXI XXXXXI xxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxx xxxx.x xxxxmxxxmm xiximmmm ■ mi mimmmm 1 1 1 1 1 I (e) 72.2 O I X X X X X I X X X X X X X XX X X XX X X XX X X XX X X XX X X XXX XX X XXI XX XX XXXX XX XIX xxxx XXXXXI xxxii mill urn xxxxxx ixxxx ximi xxmiixmi imimixxx xxmmmi mxiimxxx IXXIXIXXXXXX I XXXXXXXXXXIX X iixxxxxxmmx imxxmxxxiii X xxmmxmm xmmmxxiix IX xxmmxmm ii xxmmxxmmxx ixxxxmixmimi iiixmxiimxxm ixxxixmxmixiix xxiximixxmim x immximmiix i ixxxmmmxxm i xmmmmxmi i i immmmmxm x xxxx mxxxxx xxxxxx mi mi xx mmmx xi i m immiiimxxmmx xmmiixmmmm xxxx mxixiimm xxxx xmixmmmmxxxx xximmmixmiixxx 1 5 10 15 20 25 i X (b) 172 .1 -u X i X i 1 X X X X X X I X II I I X I XI XX I I I X IXI X XI X XXX XIII XXXI X III I IXI IIII xm I III I X XIX I IXXK XXXX mix x XXXI mn i XXII nix mix i mix i nix ixxx mil i urn n xm immx in mi immxxim xmmmxii ■immiiim i mxxmxmmm tmxxmmixmx mil XXXII IXXXX 11111 IIXXI mix XXIII nmiixmx mxxxxxmi mnmim immmii mimmmxixm mimiixxxxiinxi nxxmxmmmxx imimmimim X X (Iiiiimnmmmix 1 1 1 1 15 20 25 (f ) 171.6 1, IIIIIXXI II XX iimiiiii xm mmmx mi ■mmmxi mi xmmmmi mimmmmim xx ixxxmmmxxm xmmimmmm ximximm IIIIXXXXXXXXX mmxxmxx II11IIIIIII I (C) 171.9 ^ I I i I X X I X I II II I X I II II II 1 II m m I XXI IXIX xm 1 III mi I XII 1 im mn i IIII mum mi mi mi IIXXI mil inn mix mum limiii imiiii limiii iixmmmm mn urn "in xxm IIXXI XXXXX III I X xxxxx mil mn ixxxx XXXII xxm XXXIII iimiinm iiiiiimm imiiimii miimim mimmii miimim x iimxiiim i immnm i X X I x miimim i immmii i rh xx ii mi m xxx xm 1 1 XI 10 15 20 25 X eg) 170.5 1^ I I I X I I > XX i ii i n I XX I II XI II XI IX ii ■linn mini XX II mini i II XI imimi II miiiini mxmiii XX XI i i iimiiiii IXIXXXXXXI XI XX I I (iiimmm iimmmn n II (iiimmm IXXXIIIIIIIII IX ii (imiim minim mimm iimiiiii iimiiiii IXI XIX XIX "J ■ II xxx .. 1 ■ II mimm XXX II iimmii XXXXXIII1I xmxmii mimm iimmii iimmii mimm IIII nil xxxx ■11 III III III III III «» III 1 1 lie" 1 1 10 15 20 25 15 20 25 (d) 170.2 a. xx mn iixximmi minimum mimmii mmm 20 (h) 172.3 O ■ I i I X i X X X X X XI I II I II II 11 XI II IXX II III III III III XXXXXXI mini imiiii i imiim i mmm i imiim x mimm i iimmii i mimm x xminiii x imiimmii minimum IIIIXXIIIIIIIIII xiiimmmiii xxxiiiimiimiii iimmmimm imimmimiii iimmmimm mimimimm nnmimimm mimimimm immmiiimxi imimiiiiiiixix nxiiiixmxuxxxsxiixju _ mmmimmnxxi mmmimmimi xmimmiimmii imxxiimmmmx mm mini mmm 10 15 20 25 HARMONIC NUMBER Fig. 3 — Set of Fourier spectra for digit simulation No. 1 (as shown in Fig. 2). Each X represents a 1-db relative amplitude increment. Spectra are in alpha- betical order with respect to time. 2842 DIGIT-SIMULATING SPEECH SOUNDS 2843 1 CL) 171.5 'V X I X 1 » 1 X I XX IX XX XX XX IX XI II m ii XXXXIX XXXXIX IIIIIXIX xixiixxxx iixiixiii X ■ IIIII1IIII "ilium xx ■IXII1IIII IIIIXIXXXX IIIIIIIXII IIIIIIIIII IIIIIIIIII XX XX XX X XX X ■ I 1 IX IX ■IIIIIIIII! ■IIIIIIIIII minimi IIIIIUIXII XXXII immim milium iiiiiiiim ■ m.mm iimimii im.imu milium nun IIIIIII mm mm -, IIIIIII ■IIIIIIIIII XXXXIIXXIXX XXXXXIXXXXX iimimii mmmu XXXXXIXXXXX IXXXIIXXXIX XXXXXIXXXXX XXIXXX X mm i XIIXII I X mm i xi mm i ii IIIIIIXXIII mmmu t 5 10 15 20 25 J j) 170.1 'V/ x X X X I I X 1 XX X XX X XX XX XX XX XX IX XX IX XXX XX xxxxxx XXXXIX X XXIXXXX X XXXXXXX X XXXXXXXXX xxxxxxxxx X i IIIIIIXXXI IXXXXXXXXX ■ IIIIIIIIII IX X IIIIIIIIII III 1 iixiiiiiii xxxxx xxxxx IXXXXIIXXI urn XXXXXXX XXXXXXX XIXIXXXXXX UIIIIIIU XXXXXXXXXXI XXXIIIXXXXI IIIIXXXXXXI immim XXXXXII milium milium IIIIIXX IXIXIII J_ xiiiiiiii XXXXXXXXXXI IIUUIIXII mum XIIIIIIX X XI IIIIIIXXXXIIIIIXIXX mmummum IIXXIIXXXXXXXXXIIXI IIIIIIIXXIIIIIXIIII IIIIIIIIII I1IIUU ■ II 1 1 1 1 5 10 1 I 15 20 25 HARMONIC NUMBER ' Fig. 3 — (continued) two sets of spectra corresponding to the two speech segments whose time domain waveforms appear in Fig. 2. V. DISCUSSION OF RESULTS Several aspects of the spectra shown in Figs. 3 and 4 are worthy of note. First of all, it is seen that these two speech segments (as well as the twelve others not shown here) do indeed satisfy the two properties anticipated in the introduction. The high degree of periodicity of these speech waveforms is spectrally confirmed by noting that in both se- quences of spectra the harmonic structure remains extraordinarily uniform. (This result also confirms, by hindsight, the original validity of a period-by-period Fourier analysis.) By noting the fundamental pitch (thus the period) of each segment, it is seen that this highly stable harmonic structure is maintained for at least the 23 milliseconds which coincides with the duration requirements of the TOUCH-TONE re- ceiver. J (a) 232.6^ x < < X X X X X X X XXX XXX XXX xxxx xxxxx III" ( xxxxx x < xxxxxx X « XXXXXX XX I XXXXXX XX ( XXXXXXXX XX X IXXXXXXXXX XX X IXXXXXXXXX XX X IXXXXXXXXXXXX X IXXXXXXXXXXXX X IXXXXXXXXXXXX X IXXXXXXXXXXXXXX (xxxxxxxxxxxxxx IXXXXXXXXXXXXXXX txxxxxxxxxxxxxxx IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX [XXXXXXXXXIIIIII IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX IXXXXXXXXXXXXXXX mxmxxxxxxxxx IXXXXXXXXXXXXXXX III 1 ; „ (e) 235. 2^ X X X X X X X X X X X X X < I X XXX X XXX I XXXXX X X XXXXX XI X XXXXX XI X XXIIXI XX I XIIIIIIIIIX I X IIIXXIIIIII I IXXXXXXIXIXIXXI XXXXIXXIIXIXIXX xixxxxxxxxxxxxx IXXXXXXXXXXXXXXX XXXXXXXXXXXXXIIX Txxxxxxxxxxxxxxx XIXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXXI IXXXXXXXXX X X XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX IIXIIXXXIIIIIIII IXXIIXXIIXXIIIXX 1 1 1 1 1 I 5 10 15 20 25 » (b)234.7'\j X X X X X X X X XII III III III XXX xxxxx X IIIII X mix I i xxiiii X XXXXXX II I XXIIXI I XXXIII II I X XIIXXX II I X XXXXXXIX X XXXXXXXX II I X XXXXXXXI X XXXXXXXX II x IXXXI X XIIIIII) IXXXX I IIIIIIIIIIIII I XXIIXXXXXXXXX XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX xxxxxxxxxi IXIIIIIXXX IXIIII XXXXXX XXXXXXXXXXXXXXXX IIIIIIIIII IXXXXX XXXXXXXXXI XXXXXXXXIX XXIXXIXXIX IIIIIIIIII XXXXXX XXXXXXXXXI XXXIXXXXXX XXXXXX xxxxxx XXXXXXXXXXXXXXXX XXXXXIXXIXIXXXIX XIIIIIIIXIIIIIII XXXXXXXXXXXXXXXX Iiixxxxxxa '"III 1 1 1 1 1 1 10 15 20 25 I 15 20 25 (f) 237. 4 O IXXXI II IXXXXXXXXX xxxxxxxxxx (0 235.30 xxxxxxxxxx -i — i — r 10 15 20 25 (g) 236.50. I I X I X X X X X X X X X X XXX X III X XII I I XIII X I XIII X X XXXXXX XX X XXXXXX XX X IXXIII XX X i mux II I IIIXXXXXI XX I XXII x I XX X XX X XXXIIXXIX XI I XX X xiixxxiiii mi xnxxxxm XXXX XXXIIIIIIX xxxx TxiTiiiiii xxxx XIIXXIIIIX xxxx X XXX I ■ XXXI XIII XXXXXXXXXX xxxx I XXXXXXXX! xxxx minim xxxx XXXXXXXXXX XXXI XXXXXXXXXX XXXI XXXXXXXXII IXXXX Iiixxxxxxa mil xxxxxxxxxx [XXXX xxxxx xxxxxxxxxx xmimiiixxii IXXXXXXXXX IXXXX IXXXXXXXXXXXXXX xiimxixi IXXXI IXXXX xxxxxxxxxx XXXXX 15 20 25 HARMONIC NUMBER 15 20 25 x (d) 236.9 <\j X X X X I X X X XII III xxx XXI X I XIIII " i mil XI X XXXXXX XX x mm XI I IX I i xxxxxx XXXXXXXX III X XXIIIXXX XXX X XXXXXXXX XXXXX mum xxm xxmm XXXXXX mum mux IIIIXI X I XX x IIXI XXXXXXXX XXXXXX IIXIIX x X XI I X1IIX I, IX. ximiiimmx XXXXXXXX XXXXXXXI XXXXIIXX XXXXXXXI XXXXXXXX XXXXXX mm II Ilium tO 15 20 25 X X I CW 236.6 X X X X X X I I X X I I I I I I I 'I xxx X XII X X I III X I xxxx X XI X xxxxx II I IIIII II I IIIII II I XXIIXI XX I xxxx** XI I i nun II I XXXXXXXX II 1 I XXXI XXXXXXXX nun mum nun mm ilium mm XXXXXXXX XXXXXX xmmmxxiii XXXXXXXXXXXXXXXX xiiimmiimi xiixxmixxxxxxx XXXXXXXXXXXXXXXX XXXXXXXXXXXXXXXX XXIXXIXXXXXXXXXX IXXXXXXXXXXXXXXX xxxxmmmm XXXXXXXX xxxxxxx mxmiiiiixm XXXXXIIXXXXXXXXI XXXXXXXXXXXXXXXX xxximiimim XXXXXXXXXXXXXXXX I l 15 20 25 Fig. 4 — Set of Fourier spectra for digit simulation No. 2 (as shown in Fig. 2). 2844 (L) 236.80. u.nuunn 15 20 25 I . » X (m) 241.2<\j X I X X X X X X X X X I X X X X X X X X X XXX X XXX X XXX X XXX I XXX X XXX X XXX X i xxxx x x X XXXI X X X XIXX X XX XXXXXX X XX XXXXXX X XX XXXXXX X XX XXXXXX X xxxx xxxxxxxx xxxxxxxx xxxx xxxxx xxxxxxxx xxxxx xxxxxxxx xxxxx xxxxxxxx xxxxx xxxxxxxx xxxxx xxxxxxxx xxxxx xxxxxxxx xxxxx xxxxxxxx xxxxx XXXIIIXXXXXXXX XXIXXXXXXXXXXXX xxxxxxxxxxxxxxx xxxxxxxx xxxxxxxx XXXXXX XXXXXX XXIXXXIIXXXIXXX xxxxxxxxxxxxxxx xxxxxxxxxxxxxxx xxxxxxxxxxxxxxx IXIIIXXXXXXXXXX XIXIXIIIIIIIXII xxxxxxxx xxxxxxxx XXXXXX xxxxxxxx (XXXXX xxxxxxxxxxxxxxx xxxxxxxxxxxxxxx 1 1 1 10 15 20 25 (j) 239.3 -Aj 10 15 20 25 X X (n) 241. 40i X X X I X < X X X X I X X < X X X X XXI XII XXI XIX X XXI X XII X XX" X I III I I II" I X XXXI X X XXXI X I X IXXX I I X XXXI X II I III! I XX I XXIXII XI I IIIXII I "11" IX XX xxxxxxxx I XIX IIXIXIIX X XXX xxxxxxxx X XXX xxxxxxxx X XXX xxxxxxxx X XXX xxxxxxxx X XXX xxxxxxxx mil xxxxxxxx mxxi XXXXXX XIIXXXII XXXXXX mlffm xxxxxx XXXXXX XIIIIXII IXXIII IIIIIXIIIXIIIII xxxxxxxx XXXXXX xxxxxxxx XXXI XI mm minimum XIIII1XIIIIXXII XXIIIIIIIXIIIII IXXXXIXI 1. ..... , mm IIXIXIIX mm xxxxxxxxxxxxxxx , * (K)239.10i x \ X X I X X X X I X X XIX XXX I III \ x mi x x i mi i i XXXXXX I I mm i x xxxxxxxx XX XXXXXXXX XX XXXXXXXX II IXXXXIXI XXI XXXXXXXX III ixxxxxxx mx xxxxxm im IXXXXXXX xxxxx ixxxxxix mm xmmi mm xixxxiix mm iixxiiii mm ixxxxnx mm XXXXXXXX XXXXXX ixxiii mm xmmimiiii xmmmmii xxxxxxxxxxxxxxx IXIIIXXXIIIIIII Ilillllllliim IIIIIXIIIXIIIII imxxxxxxxxixx xmmmmix , J (l) 239.10; i mx x x ml I i mi x x XIIIIIIX I XXI IIIIIIII X III mmiiiiixxi ■ ximiixxxxixx "lIIIIXIXXIXXX IIIIIIIIIIIIIII XXIXIXIXXIXXXIX Imimixxxm mmmxmxx Illlliuiulm immmiixii iiimmilim IIIIIIXIIIXIXII IIIIIIIIIIIIIII IIIXIIIIIXIIIXI 15 20 25 10 15 20 25 X X (0) 240.8O1 X X X X X X X I X X X X X X III X XXX X III i mi J mm X iiiimi xxxxxxxx xxxxxxxx X J xxxxxxxx XIIIIIIX X II XXXXXXXX X XI IXXXXIIX IIIIIIII I II XXIXXXXX XXXXXXXX IXIIII xxxxxxxx XXXXXX XXXXXXXX XXXXXX IXXXXXXX ""x" IIIXXXXX IXXXII IIIIIIII IXXXXXXX XXXXII IIIIIIII IIIXII IIXXIIII IIXXII IXXIII XXXXXXXX IIIIIIII xxixiimimn IXXXXIIXXXXXXXX xximxxmmi 15 20 25 15 HARMONIC NUMBER 10 15 20 25 Fig. 4 — (continued) 2845 2846 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1963 Secondly, one finds immediate justification for the fact that these speech segments caused digit simulation. By multiplying the funda- mental pitch of any segment by the orders of its two dominant har- monics, a valid TOUCH-TONE calling signal* is derived. Thus a voice- produced digit simulation is spectrally analogous to a valid TOUCH- TONE signal accompanied by noise, with the sole exception that in the former case both the "noise" and signal components are integral mul- tiples of a discrete fundamental frequency. Indeed, this sole distinction between a digit simulation and a valid signal might possibly be used to provide further simulation protection in future voice-frequency signal- ing applications. Specifically, a receiver might be designed to be sensitive to the presence of selected harmonics and/or sub-harmonics of valid signal frequencies, and thereby to reject many speech phonemes which would ordinarily cause simulation. In the portions of Figs. 3 and 4 where the harmonic structure is notice- ably changing with time (namely at the beginning and end of each series of spectra) pitch-synchronous Fourier analysis can be regarded as only an approximation of spectral density. For some applications, however, the approximation is still useful. In the first place, one can obtain a practical "feel" for the rate of change of pitch and harmonic structure in vowel- type speech sounds. Also, from the standpoint of digit simula- tion, by examining the spectra one can ascertain just how and when a speech segment becomes a digit simulation. For example, in the early spectra of Fig. 3, although pitch requirements for digit simulation are satisfied, the 10th harmonic competes with the 7th harmonic for limiter capture, and receiver recognition is prevented by limiter guard action (i.e., insufficient signal-to-noise ratio). (See Ref. 1, pp. 10-11, 13.) On the other hand, although early spectra of Fig. 4 show an acceptable harmonic structure for digit simulation, the pitch is slightly too low for receiver recognition. In a similar manner, one can determine how and when a digit simulating wave-form starts to degenerate. Admittedly, the speech segments chosen here are both rare and few in number. Thus, one cannot draw conclusions of statistical significance from this study. However, there is no reason to believe that any other group of frequencies of the same capacity in the voice band would not be simulated by the voice about as often as were these TOUCH-TONE calling frequencies. Therefore, such vowel-type speech segments may be * A valid TOUCH-TONE calling signal consists of one frequency from each of two groups: a low group — 697, 770, 852 and 941 cps ±2.5 per cent — and a high group — 1209, 1336, 1477 and 1633 cps ±2.5 per cent. DIGIT-SIMULATING SPEECH SOUNDS 2847 looked upon as potential digit simulations in almost any proposed voice- frequency signaling application. REFERENCES 1. Battista, R. N., Morrison, C. G., and Nash, D. H., Signaling System and Re- ceiver for TOUCH-TONE Calling, Trans. IEEE, 66, March, 1963. 2. Mathews, M. V., Miller, Joan E., and David, E. E., Pitch Synchronous Analysis of Voiced Sounds, J. Acoust. Soc. Am., 33, Feb., 1961, pp. 179-186.